An Egf Signaling Map in Pathway Logic

Size: px

Start display at page:

Download "An Egf Signaling Map in Pathway Logic"

Aron Parker
5 years ago
Views:

1 An Egf Signaling Map in Pathway Logic M. Knapp, L. Briesemeister, S. Eker, P. Lincoln, I. Mason, C. Talcott, and K. Laderoute SRI International Menlo Park, CA Panel 1

2 Abstract Pathway Logic is an approach to modeling cellular processes based on rewriting logic, a simple logic designed for modeling and analysis of distributed systems. It allows one to model aspects of the structure and state of interacting components as elements of an abstract data type; to represent individual process steps (reactions) as rewrite rules; and to study possible ways a system might evolve using techniques based on logical inference. Given a network of reactions and a specification of cellular components one can query the network about possible reaction pathways and outcomes. Knockouts that prevent a given outcome can be computed, competing reactions can be found, and pathways can be compared to look for potential cross-talk. Epidermal growth factor receptor (EgfR) signaling regulates growth, survival, proliferation, and differentiation in mammalian cells. We present a Pathway Logic model of early response to Epidermal growth factor (Egf) stimulation in adherent cells expressing Egf-receptors. The model is entirely based on experimental results and data curated from the published scientific literature. We explain the curation process that extracts information about state changes from experimental data to determine the components of a reaction rule. A reaction network assembled from data supporting events that might be downstream of EgfR signaling has over 370 reactions involving more than 460 species (signaling molecules in different states and locations). The network was then constrained to outcomes that have been demonstrated experimentally to happen in response to a short stimulus with Egf. So far more than 80 such outcomes have been collected and used to test the adequacy of the model---does it predict the observed outcomes? We will also point out some surprises that appear in the generated pathways due to the richness of the considered biological context. Pathway Logic models are qualitative and thus answer different kinds of questions than quantitative differential equation-based or stochastic models. They can also serve as useful road maps for developing quantitative models of subsystems and understanding the potential cross talk between signaling submodules. Panel 2

3 Introduction Various representations or models of Egf-to-Erk signal transduction have been described over the last decade. In general, this process is modeled as follows: Egf EgfR Grb2 Sos1 a Ras family member Raf1 MEK1/2 ERK1/2 In this canonical pathway, Egf binds to the EGF receptor (EgfR) and stimulates its protein tyrosine kinase activity to cause autophosphorylation. Next, a complex containing the adaptor protein Grb2 and the guanine nucleotide exchange factor Sos1 docks (binds) to the autophosphorylated (activated) EgfR. The Sos1-containing EgfR complex activates a Ras family GTPase, and the activated Ras protein activates Raf1, a member of the RAF serine/threonine protein kinase family. Raf1 then activates the dual-specificity protein kinases Mek1 and/or Mek2 (MEK1/2), which activate Erk1 and/or Erk2 (ERK1/2). Here we present a Pathway Logic model of early signal transduction by EgfR that is entirely based on experimental results curated from the published scientific literature. We demonstrate that the series of events between activation of EgfR by Egf and activation of Erk2 may not be as simple as those described in the canonical pathway. A note about protein nomenclature: the symbols used for proteins in this presentation are those used in Pathway Logic. Other identifiers and synonyms are provided in the Glossary in the Appendix. Panel 3

4 Data Collection The first step in the construction of the model was to collect the data: 174 papers were searched for appropriate experiments and the results were listed as 1373 evidence items. Below is an example of an item used as evidence for rule E19 which requires the presence of tyrosine phosphorylated Gab1 for the activation of Erk2 in response to Egf. An Evidence Item Source PMID: Type: data Figure : 2 Subject Pathway Logic name: Erk2 Expressed: yes Identification method : expression tag antibody State Change Type: kinase activity Direction: increase Assay: IP Kinase assay, MBP as substrate Cause Requirement Environment Stimulus: Egf Time: 5 minutes Concentration: 0.25 ng/ml Pathway Logic name: Gab1 Method: Y627F dominant-negative coexpression Method: Y659F dominant-negative coexpression Cells: COS-7 State: serum starved for 20 hr Evidence items do not use the conclusions of the authors, only the experimental results. Panel 4

5 Writing the Rules Initially a set of Pathway Logic rules were derived from published pathway diagrams and review articles. The rules were annotated with evidence items that support or contradict them. If no evidence could be found to support a rule, it was either removed or modified to agree with the data. The remaining evidence items were used to construct rules not found in conventional pathway diagrams. The rules in this model are divided into two sets: (1) Common Rules are based on state changes caused by ectopically expressed proteins. (2) Egf Rules are based on state changes caused by short-term Egf treatment. In most cases, the treatment was 10 minutes or less, which is the time window corresponding to the peak response of Erk phosphorylation. Choosing an Initial State The initial state for a Pathway Logic model consists of a list of components (proteins, chemicals, or nucleic acids), their modifications, and locations. In this case, the initial state represents a serum starved, adherent cell expressing EgfR. The list was curated from published experimental data (Table I, Appendix). The initial state also contains one ligand - Egf. Panel 5

6 About Petri Nets In Pathway Logic, reaction networks and pathways are represented as Petri nets. Petri nets were invented to model execution of concurrent processes (such as signals propagating through a cell). A Petri net can be thought of as graph with two kinds of nodes: transitions/rules (reactions -- shown as squares) and places/ occurrences (reactants, products, modifiers -- shown as ovals). Each occurrence represents a chemical, protein or complex in a specific state and location. The reactants of a rule are the occurrences connected to the rule by arrows from the occurrence to the rule. The products of a rule are the occurrences connected to the rule by arrows from the rule to the occurrence. The modifiers of a rule are the occurrences connected to the rule by a dashed arrow. For example rule 2 has reactant C, product D and modifier AB. To execute a Petri net model one puts tokens on the ovals corresponding to occurrences present in the initial state (represented by darker colors). A rule can fire if all of its reactants and modifiers have tokens. When a rule fires the tokens are moved from reactants to products, indicating that the products are now present. Modifiers are unchanged. An execution is a sequence of firings of enabled rules. (Note there may several possible executions.) A pathway is a network with initial marking that corresponds to rules fired in one possible execution. C A 2 D Panel 6 1 AB B C A 2 D 1 AB B C A 2 D 1 AB B

The result was a Petri net that included all curated events that might occur in response to a short-term stimulus with EGF.

7 Building a Petri Net The rules and initial state files were loaded into the Pathway Logic Assistant (PLA). PLA converted all the rules into one Petri net and then removed any transitions not allowed by the initial state. The result was a Petri net that included all curated events that might occur in response to a short-term stimulus with EGF. At this point the Petri net looked like this: The Unconstrained Petri Net: Events that could occur in response to Egf Although PLA can be used to browse this network, it is clearly too complex to understand as a whole. Panel 7

8 Making a Subnet We were interested in the mechanism of the initial activation of Erk2 in response to a stimulus by Egf. Once a Petri net had been built, the PLA Viewer was used to make a subnet in which anything not relevant to the Erk2 activation was stripped away. Unfortunately, there are still so many potential paths to Erk2 activation that the Petri net is still too complicated to comprehend without visualization tools. A Subnet of the Unconstrained Petri Net All possible routes to from Egf to Erk2 activation Panel 8

9 Finding Paths Given an initial state and a goal state, the PLA viewer can isolate possible paths between them. It uses a model-checking tool named LoLa ( that returns the first path it finds -- usually the path with the least number of steps. We asked LoLa to find a Path from Egf to Erk2 activation. The first path that LoLa found (Path #1) is similar to the canonical pathway but uses Braf instead of Raf1 to activate Mek1 and requires activated Prkcz as well as Mek1 to activate Erk1 and Erk2. Is this the "correct" path? Unfortunately, we have no tools to answer that question automatically. But we have provided the evidence collected to date to support each rule. By inspecting the evidence for each of the rules, it becomes apparent that the evidence supporting rule 452 is weak; there is only one datum and it is an in vitro kinase assay showing that recombinant Mek1 can be activated by recombinant Braf. PLA allows the user to "hide" rule 452 and ask for another path. The new path (Path #2) uses rule 288 in which both Braf and Raf1 are activated by Hras, Mlk3, and Ywhaz. This result correlates well with the next rule E32 in which Mek1 activation in response to Egf requires Raf1, Mlk3, Gab1, and IqGap. The user can access the evidence for a rule via PLA or in the curated model. It is left up to the user to decide whether there is sufficient data to justify the rules. Panel 9

10 EgfR-CLm Egf-Out Sos1-CLc Grb2-CLc (Egf:EgfR-act)-CLm (Grb2:Sos1)-CLc E74-2 Sos1-reloc-CLi Grb2-reloc-CLi Hras-GDP-CLi 999 Pi3k-CLc Hras-GTP-CLi Braf-CLc Pi3k-act-CLi PIP2-CLm Braf-act-CLi Mek1-CLc PIP3-CLm Pdpk1-CLc 108 Mek1-act-CLi Path #1 Pdpk1-act-CLi Prkcz-CLc Erk1-CLc Erk2-CLc Prkcz-act-CLi 14 Erk2-act-CLi Erk1-act-CLi Panel 10

Panel 11 643 PIP3-CLm 111-1 Prkcz-act-CLi Sos1-reloc-CLi 999 Ywhaz-CLc 288 Braf-act-CLi Raf1-act-CLi E09-2 Crk-Yphos-CLi 53 Rac1-GTP-CLi Erk2-act-CLi Hras-GTP-CLi 325 Pi3k-act-CLi Erk1-act-CLi 108

11 Panel PIP3-CLm Prkcz-act-CLi Sos1-reloc-CLi 999 Ywhaz-CLc 288 Braf-act-CLi Raf1-act-CLi E09-2 Crk-Yphos-CLi 53 Rac1-GTP-CLi Erk2-act-CLi Hras-GTP-CLi 325 Pi3k-act-CLi Erk1-act-CLi 108 Pdpk1-act-CLi IqGap1-reloc-CLi E Mlk3-act-CLi Pdpk1-CLc E39-2 Ptk6-act-CLi E40-2 Mek1-CLc Gab1-CLc E (Egf:EgfR-act)-CLm Erk1-CLc 14 Hras-GDP-CLi Braf-CLc Raf1-CLc (Grb2:Sos1)-CLc E74-2 Grb2-reloc-CLi Rac1-GDP-CLi Prkcz-CLc Mlk3-CLc Pxn-Yphos-CLi 1092 E75-2 EgfR-CLm Sos1-CLc Grb2-CLc Egf-Out PIP2-CLm Ptk6-CLc Pi3k-CLc Gab1-Yphos-CLi Mek1-act-CLi IqGap1-CLc Crk-CLc Erk2-CLc Pxn-CLc Path #2

12 Constraining the Egf Map There are many possible paths in addition to Paths 1 and 2. One could go on evaluating each one manually or use the PLA tools to limit (or constrain) the number of possible interactions by giving precedence to events found in response to Egf. We applied two constraints to the network of Panel 7. The first constraint took advantage of a list of 85 state changes demonstrated experimentally to occur in response to a short stimulus with Egf (Table II, Appendix). These occurrences (protein states) were set as goals and a set of concurrent paths were produced by PLA. This selection ensures that the paths used to reach the chosen goals are mutually compatible (i.e., no rules would be used that would allow one goal to be reached at the expense of any other goal). The second constraint used PLA's ability to hide rules, thus giving precedence to Egf Rules over Common Rules. The Egf Rules contain requirements specific to Egf signaling that must be satisfied before they can fire. By using Egf Rules instead of Common Rules wherever possible, LoLa is forced to use events must happen before Erk2 is activated. Panel 12

13 The Constrained Egf Map The Egf Map above was constructed from the network of Panel 7 using both constraints. It is much simpler, and more focused, but still sufficiently complex that tools are needed to explore and understand it. Panel 13

14 The Path to Erk2 in the Constrained Egf Map Panel 15 shows the path from the constrained Egf map (Panel 13) that leads to activated Erk2. This differs from the path found by searching in the unconstrained network (Panel 7) as we have forced the model-checking tool to work in the context of realizing all the other observed goals. This path is a great deal more complicated than the canonical pathway. This result is not surprising if you look at all the reported requirements for Erk2 to be activated. (Table III, Appendix) contains a list of the proteins that have been tested for their influence on Erk activation and/ or phosphorylation in response to Egf. The shaded rows in the table are events that the curator chose to use in the Egf Rules. Clearly, there are many unfamiliar events within the new Egf to Erk2 path. We point out three such events in the following panels. (1) Rala is required for Src activation in response to Egf (Panel 16). (2) Sos1 is not required for Hras activation in response to Egf (Panel 17). (3) Mlk3 is required for the activation of Braf, Mek1/2, and Erk1/2 (Panel 18). Panel 14

15 A Path from Egf to Erk2 activation derived from Egf Specific Rules EgfR-CLm Egf-Out E76 Tnk2-CLc (Egf:EgfR-act)-CLm Gab1-CLc Ptk6-CLc Eps8-CLc Abi1-CLc E49 E21 E Tnk2-act-CLi RasGrf1-CLc Ptpn11-CLc Pi3k-CLc Gab1-Yphos-CLi Sos1-CLc Ptk6-act-CLi (Abi1:Eps8)-CLc Pxn-CLc 736 E62 E RasGrf1-act-CLi Ptpn11-Yphos-CLi Hras-GDP-CLi Pi3k-act-CLi Rac1-GDP-CLi (Sos1:Abi1:Eps8)-CLc Pxn-Yphos-CLi E25 E41 Rgl1-CLc Hras-GTP-CLi Rac1-GTP-CLi Mlk3-CLc Rgl1-reloc-CLi Rala-GDP-CLi Braf-CLc Mlk3-act-CLi E43 E04 Rala-GTP-CLi Src-CLi IqGap1-CLc Braf-act-CLi Raf1-CLc E54 E75 E42 Ptk2b-CLi Src-act-CLi IqGap1-reloc-CLi Mek1-CLc Raf1-act-CLi E60 E32 Ptk2b-act-CLi Erk2-CLc Erk1-CLc Mek1-act-CLi E19 Erk2-act-CLi Erk1-act-CLi Panel 15

16 EgfR-CLm E76 Egf-Out Egf to Src (rule E54) Tnk2-CLc (Egf:EgfR-act)-CLm Gab1-CLc Ptk6-CLc Eps8-CLc Abi1-CLc E49 E21 E Tnk2-act-CLi RasGrf1-CLc Ptpn11-CLc Pi3k-CLc Gab1-Yphos-CLi Sos1-CLc Ptk6-act-CLi (Abi1:Eps8)-CLc Pxn-CLc 736 E62 E RasGrf1-act-CLi Ptpn11-Yphos-CLi Hras-GDP-CLi Pi3k-act-CLi Rac1-GDP-CLi (Sos1:(Abi1:Eps8))-CLc Pxn-Yphos-CLi E25 E41 Rgl1-CLc Hras-GTP-CLi Rac1-GTP-CLi Mlk3-CLc Rgl1-reloc-CLi Rala-GDP-CLi Braf-CLc Mlk3-act-CLi E43-2 E04 Rala-GTP-CLi Src-CLi E54 Ptk2b-CLi Src-act-CLi E60 Ptk2b-act-CLi Src has IqGap1-CLcbeen Braf-act-CLi found Raf1-CLc to be activated in response to Egf in E75 E42 various cell types and the action of a tyrosine IqGap1-reloc-CLi Mek1-CLc Raf1-act-CLi phosphatase (Ptpn11) to dephosphorylate E32 Src Y527, releasing Src from an auto- Erk2-CLc Erk1-CLc Mek1-act-CLi inhibited state, has long been recognized. Yet a possible E19 requirement for Rala in the activation Erk1-act-CLi of Src [ ] has been ignored. Erk2-act-CLi Panel 16

17 EgfR-CLm E76 Egf-Out Egf to Hras (rule E25) Tnk2-CLc (Egf:EgfR-act)-CLm Gab1-CLc Ptk6-CLc Eps8-CLc Abi1-CLc E49 E21 E Tnk2-act-CLi RasGrf1-CLc Ptpn11-CLc Pi3k-CLc Gab1-Yphos-CLi Sos1-CLc Ptk6-act-CLi (Abi1:Eps8)-CLc Pxn-CLc 736 E62 E RasGrf1-act-CLi Ptpn11-Yphos-CLi Hras-GDP-CLi E25 Rgl1-CLc Hras-GTP-CLi 337 Rgl1-reloc-CLi Rala-GTP-CLi Src-CLi E54 Ptk2b-CLi Src-act-CLi E60 Ptk2b-act-CLi E43-2 (Sos1:(Abi1:Eps8))-CLc Pxn-Yphos-CLi experimental evidence E41 for only Rala-GDP-CLi Common Rules. Braf-CLc There Mlk3-act-CLi are six Erk2-CLc Pi3k-act-CLi Rac1-GDP-CLi We were able to find two proteins required for Hras Rac1-GTP-CLi Mlk3-CLc activation by Egf: Gab1 and Ptpn11. The requirement for a 470 RasGef was assumed from the RasGefs that have been shown E04 to activate Hras in the Common Rules: RasGrf1, RasGrf2, IqGap1-CLc Braf-act-CLi Raf1-CLc RasGrp1, RasGrp3, RasGrp4, and Sos1. E75 Sos1 E42is the only RasGef that has been tested to IqGap1-reloc-CLi Mek1-CLc Raf1-act-CLi see if it is required for Hras activation by Egf [ ]. E32 Surprisingly, the authors found that the increase in endogenous Erk1-CLc Mek1-act-CLi Hras-GTP after 5 minutes of Egf treatment E19 was not affected by knockout of Sos1 in mouse Erk2-act-CLi Erk1-act-CLi embryo fibroblasts. Panel 17

EgfR-CLm E76 Egf-Out Contribution of Mlk3 Tnk2-CLc (Egf:EgfR-act)-CLm Gab1-CLc Ptk6-CLc Eps8-CLc Abi1-CLc E49 E21 E39 1159 Tnk2-act-CLi RasGrf1-CLc Ptpn11-CLc Pi3k-CLc Gab1-Yphos-CLi Sos1-CLc

18 EgfR-CLm E76 Egf-Out Contribution of Mlk3 Tnk2-CLc (Egf:EgfR-act)-CLm Gab1-CLc Ptk6-CLc Eps8-CLc Abi1-CLc E49 E21 E Tnk2-act-CLi RasGrf1-CLc Ptpn11-CLc Pi3k-CLc Gab1-Yphos-CLi Sos1-CLc Ptk6-act-CLi (Abi1:Eps8)-CLc Pxn-CLc 736 E62 E RasGrf1-act-CLi Ptpn11-Yphos-CLi Hras-GDP-CLi Pi3k-act-CLi Rac1-GDP-CLi (Sos1:(Abi1:Eps8))-CLc Pxn-Yphos-CLi E25 E41 Another protein that is Rgl1-CLc not usually associated with the Egf signaling pathway is Mlk3. Yet Mlk3 has been reported to be required for activation of Braf, Erk1, Erk2, Rala-GTP-CLi Mek1, and Mek2 Hras-GTP-CLi 337 Src-CLi [ , ] E54 in response to Egf. Ptk2b-CLi Rgl1-reloc-CLi Src-act-CLi E43-2 Rala-GDP-CLi IqGap1-CLc E75 IqGap1-reloc-CLi Rac1-GTP-CLi Mlk3-CLc 470 Braf-CLc Mlk3-act-CLi E04 Braf-act-CLi Raf1-CLc E42 Mek1-CLc Raf1-act-CLi E60 E32 Ptk2b-act-CLi Erk2-CLc Erk1-CLc Mek1-act-CLi E19 Erk2-act-CLi Erk1-act-CLi Panel 18

19 Validation of the Egf Map The most obvious way to validate our model is by experimentation. The data used to prepare this map was obtained from published experiments from different research groups using different cell lines, times of Egf stimulation, levels of EgfR saturation, lysis methods, antibodies, plasmids, and knockout technologies. The analysis used to generate our model assumes that variations in the assays and materials did not significantly change the conclusions used to generate the rules. This major assumption could be tested by reproducing the Egf to Erk experiments with a standardized cell line and protocol. Conclusion The Pathway Logic model of Egf stimulation was developed by curation of a knowledge base of signaling reactions from experimental data in the published literature. Both general reactions and reactions specific to cells stimulated by Egf have been curated. Subnets and paths were assembled by postulating an initial state (proteins expressed in a cell and external ligands) and using logical tools to: (1) find all reactions that might be reached given the intial state (2) find a pathway realizing specified observations (occurrences) within the full network, or within subnetworks. This often leads to pathways that are more complex than the usual predefined pathways. These inferred pathways can serve as a guide for developing dynamic models as well as to suggest further experimental studies. Panel 19

20 References [ ] Goi T, Shipitsin M, Lu Z, Foster DA, Klinz SG, Feig LA. An EGF receptor/ral-gtpase signaling cascade regulates c-src activity and substrate specificity. Embo J 2000;19: [ ] Qian X, Esteban L, Vass WC, Upadhyaya C, Papageorge AG, Yienger K, Ward JM, Lowy DR, Santos E. The Sos1 and Sos2 Ras-specific exchange factors: differences in placental expression and signaling properties. Embo J 2000;19: [ ] Chadee DN, Kyriakis JM. MLK3 is required for mitogen activation of B-Raf, ERK and cell proliferation. Nat Cell Biol 2004;6: [ ] Chadee DN, Xu D, Hung G, Andalibi A, Lim DJ, Luo Z, Gutmann DH, Kyriakis JM. Mixed-lineage kinase 3 regulates B-Raf through maintenance of the B-Raf/Raf-1 complex and inhibition by the NF2 tumor suppressor protein. Proc Natl Acad Sci U S A 2006;103: Find out more about Pathway Logic at: This work was supported by grants GM and CA from the National Institutes of Health. Panel 20

Applications of Rewriting Logic in Biology

Applications of Rewriting Logic in Biology iv UsinG ThE Pathway Logic Assistant Carolyn Talcott SRI International July 2007 The SMAll KB Live A Model of EGF STIMULation The Canonical Egf-Erk Pathway Egf